When “Less Is More” – A Critique of the Heuristics & Biases Approach to Judgment and Decision Making Psychology 466: Judgment & Decision Making Instructor:

Slides:



Advertisements
Similar presentations
Experiments and Variables
Advertisements

Correlation & Regression Chapter 15. Correlation statistical technique that is used to measure and describe a relationship between two variables (X and.
Robust decision making in uncertain environments Henry Brighton.
Additional Topics in Regression Analysis
Chapter 7 Correlational Research Gay, Mills, and Airasian
Educational Research: Correlational Studies EDU 8603 Educational Research Richard M. Jacobs, OSA, Ph.D.
Relationships Among Variables
1 1 Slide © 2008 Thomson South-Western. All Rights Reserved Slides by JOHN LOUCKS & Updated by SPIROS VELIANITIS.
Statistical Analyses & Threats to Validity
1 Least squares procedure Inference for least squares lines Simple Linear Regression.
Understanding Statistics
STA Lecture 161 STA 291 Lecture 16 Normal distributions: ( mean and SD ) use table or web page. The sampling distribution of and are both (approximately)
Educational Research: Competencies for Analysis and Application, 9 th edition. Gay, Mills, & Airasian © 2009 Pearson Education, Inc. All rights reserved.
Chapter 1 Introduction to Statistics. Statistical Methods Were developed to serve a purpose Were developed to serve a purpose The purpose for each statistical.
CHAPTER 17: Tests of Significance: The Basics
Properties of OLS How Reliable is OLS?. Learning Objectives 1.Review of the idea that the OLS estimator is a random variable 2.How do we judge the quality.
Chapter 7 Sampling Distributions Statistics for Business (Env) 1.
CROSS-VALIDATION AND MODEL SELECTION Many Slides are from: Dr. Thomas Jensen -Expedia.com and Prof. Olga Veksler - CS Learning and Computer Vision.
Robust Estimators.
Experimentation in Computer Science (Part 2). Experimentation in Software Engineering --- Outline  Empirical Strategies  Measurement  Experiment Process.
Psychology of Judgment & Decision Making: Future Directions Psychology 466: Judgment & Decision Making Instructor: John Miyamoto 12/10/2015: Lecture 11-2.
Linear Judgment Models: What Do They Suggest About Human Judgment?
When “Less Is More” – A Critique of the Heuristics & Biases Approach to Judgment and Decision Making Psychology 466: Judgment & Decision Making Instructor:
Classification Ensemble Methods 1
Machine Learning 5. Parametric Methods.
Bayes Theorem, a.k.a. Bayes Rule
Markov-Chain-Monte-Carlo (MCMC) & The Metropolis-Hastings Algorithm P548: Intro Bayesian Stats with Psych Applications Instructor: John Miyamoto 01/19/2016:
Tree and Forest Classification and Regression Tree Bagging of trees Boosting trees Random Forest.
Overfitting, Bias/Variance tradeoff. 2 Content of the presentation Bias and variance definitions Parameters that influence bias and variance Bias and.
The Psychology of Inductive Inference Psychology 355: Cognitive Psychology Instructor: John Miyamoto 5/26/2016: Lecture 09-4 Note: This Powerpoint presentation.
1 Machine Learning Lecture 8: Ensemble Methods Moshe Koppel Slides adapted from Raymond J. Mooney and others.
The Representativeness Heuristic then: Risk Attitude and Framing Effects Psychology 355: Cognitive Psychology Instructor: John Miyamoto 6/1/2016: Lecture.
Inferential Statistics Psych 231: Research Methods in Psychology.
Computing with R & Bayesian Statistical Inference P548: Intro Bayesian Stats with Psych Applications Instructor: John Miyamoto 01/11/2016: Lecture 02-1.
Methods of Presenting and Interpreting Information Class 9.
Statistical analysis.
Chapter 8: Estimating with Confidence
Chapter 8: Estimating with Confidence
Making inferences from collected data involve two possible tasks:
Statistical analysis.
PCB 3043L - General Ecology Data Analysis.
Understanding Results
Statistical Analyses & Threats to Validity
When “Less Is More” – A Critique of the Heuristics & Biases Approach to Judgment and Decision Making Psychology 466: Judgment & Decision Making Instructor:
1st: Representativeness Heuristic and Conjunction Errors 2nd: Risk Attitude and Framing Effects Psychology 355:
Four Linear Judgment Models
Psychology 466: Judgment & Decision Making
Slides by JOHN LOUCKS St. Edward’s University.
Judgment & Decision Making A Few Final Comments
Educational Research: Correlational Studies
Brief History of Cognitive Psychology
My Office Hours I will stay after class on both Monday and Wednesday, i.e., 1:30 Mon/Wed in MGH 030. Can everyone stay if they need to? Psych 548, Miyamoto,
Introduction to Bayesian Model Comparison
Linear Model Selection and regularization
Chapter 8: Estimating with Confidence
Set Up for Instructor Classroom Support Services (CSS), 35 Kane Hall,
Product moment correlation
Inferential Statistics
15.1 The Role of Statistics in the Research Process
Chapter 8: Estimating with Confidence
Chapter 8: Estimating with Confidence
Chapter 8: Estimating with Confidence
Set Up for Instructor Classroom Support Services (CSS), 35 Kane Hall,
Chapter 8: Estimating with Confidence
Chapter 8: Estimating with Confidence
Chapter 8: Estimating with Confidence
Chapter 8: Estimating with Confidence
Chapter 8: Estimating with Confidence
Chapter 8: Estimating with Confidence
Chapter 8: Estimating with Confidence
Presentation transcript:

When “Less Is More” – A Critique of the Heuristics & Biases Approach to Judgment and Decision Making Psychology 466: Judgment & Decision Making Instructor: John Miyamoto 11/01/2016: Lecture 06-1 Note: This Powerpoint presentation may contain macros that I wrote to help me create the slides. The macros aren’t needed to view the slides. You can disable or delete the macros without any change to the presentation.

Misconceptions promoted by the heuristics and biases movement. Accuracy/Effort Tradeoff – ♦ Is it a valid description of decision making practice? ♦ When is it valid? Less-is-more: Bias/Variance Tradeoff – why less-is-more Opposing Views of JDM: Heuristics & Biases Research Ecological Rationality Naturalistic Decision Making Outline Psych 466, Miyamoto, Aut '16 2 Intro to Gigerenzer & Brighton (2009) Paper

Gigerenzer & Brighton (2009) Gigerenzer, G., & Brighton, H. (2009). Homo Heuristicus: Why biased minds make better inferences. Topics in Cognitive Science, 1, GB: Abbreviation for Gigerenzer & Brighton (2009) HB: Abbreviation for heuristics & biases movement. Psych 466, Miyamoto, Aut '16 3 Meaning of "Heuristic"

"The term heuristic is of Greek origin, meaning ‘‘serving to find out or discover.’’ (Gigerenzer & Brighton, p. 108) "A heuristic technique, often called simply a heuristic, is any approach to problem solving, learning, or discovery that employs a practical method not guaranteed to be optimal or perfect, but sufficient for the immediate goals." (Wikipedia: Psych 466, Miyamoto, Aut '16 4 Three Misconceptions that were Promoted by the Heuristics and Biases Movement

Three Misconceptions Promoted by HB Movement GB (p. 109) : The HB movement gave rise to three misconceptions about human reasoning: 1.Heuristics are always second-best. 2.We use heuristics only because of our cognitive limitations. 3.More information, more computation, and more time would always be better. Hypothesis 3 implies that there is an Accuracy/Effort Tradeoff: "If you invest less effort, the cost is lower accuracy." GB, p "Less Is More" Effect: In some situations, less information (less effort) leads to greater accuracy. ♦ Contradicts that accuracy and effort are always inversely correlated. Psych 466, Miyamoto, Aut '16 5 Diagram Depicting the Accuracy/Effort Tradeoff

Psych 466, Miyamoto, Aut '16 6 Accuracy-Effort Tradeoff Method 1: Multiple regression applied to existing data. Method 2: Multiple regression applied to a judge’s predictions Method 3: Importance weighting method Method 4: Unit weighting model Method 0: Intuitive judgment (holistic judgment). Same Slide with Note re Less-Is-More Effect Accuracy High Effort High Accuracy Low Effort Low

Psych 466, Miyamoto, Aut '16 7 Accuracy-Effort Tradeoff Method 1: Multiple regression applied to existing data. Method 2: Multiple regression applied to a judge’s predictions Method 3: Importance weighting method Method 4: Unit weighting model Method 0: Intuitive judgment (holistic judgment). Reminder re Importance of Linear Judgment Models for Study of JDM Accuracy High Effort High Accuracy Low Effort Low Less-Is-More Effect contradicts the claim that there is an accuracy-effort tradeoff. Sometimes less effort produces greater accuracy (Gigerenzer & ABC Group)

Importance of Linear Judgment Models in JDM 1960 – 1980: Provided strong evidence that human judgment had difficulty processing complex information. ♦ Gave momentum to the heuristics & biases movement – present: Continued study of linear judgment models provides evidence for the value of heuristic reasoning. ♦ Example: The less-is-more effect. Gigerenzer & Brighton (2009). When sample sizes are small, unit weighting (tallying) model out-performs multiple regression. Why? Because regression weights have high variance when sample sizes are small. Psych 466, Miyamoto, Aut '16 8 Define Less-Is-More Effects

Less-Is-More Effects Decision behavior in a given environment (context) exhibits a Less-Is-More (L-I-M) effect.... if a simpler, lower-information computation can outperform a more complex and theoretically better computation in this environment. Psych 466,, Miyamoto, Aut '16 9 How to Demonstrate that Less-Is-More Effects Occur in Decision Making

How to Demonstrate that Less-Is-More Effects Occur in Decision Making? Compare simple heuristic model versus complex "rational" model Method of cross-validation can be used to determine whether the simple heuristic model or the complex "rational" model has greater predictive accuracy in a given situation (environment). (Method of cross-validation explained in the next slides) Psych 466,, Miyamoto, Aut '16 10 Diagram Explaining the Method of Cross-Validation

50% of data in test sample: Evaluate goodness of fit of predictions on the test sample Cross-Validation Studies For Evaluating the Predictive Accuracy of a Model Use a computer to repeat the procedure over many random splits. Predictive accuracy is the accuracy of prediction over many random splits. Psych 466, Miyamoto, Aut '16 11 Sample of Data random split 50% of data in estimation sample: Estimate parameters of the model Predict Same Slide Without Blocking Rectangles

50% of data in test sample: Evaluate goodness of fit of predictions on the test sample Cross-Validation Studies For Evaluating the Predictive Accuracy of a Model Use a computer to repeat the procedure over many random splits. Predictive accuracy is the accuracy of prediction over many random splits. Psych 466, Miyamoto, Aut '16 12 Sample of Data random split 50% of data in estimation sample: Estimate parameters of the model Predict Transition to a Discussion of the Tallying Model

Next: Tallying versus Multiple Regression Explain Tallying Explain methodology of comparing tallying to multiple regression Psych 466,, Miyamoto, Aut '16 13 Describe the Tallying Model

Tallying (a.k.a. Unit Weighting or Equal Weighting) When all of the cues are dichotomous, i.e., present or absent, then tallying is the same is counting the number of positive cues and subtracting the number of negative cues. When at least some of the cues are continuous like height or income, then the cues should be converted to z-scores before computing the predicted z-scores. ♦ After computing the predicted z-scores, it is necessary to convert the predicted z-scores back to the original scale Tallying is a simple heuristic model. Compare it to multiple regression, a complex "rational" model. Psych 466, Miyamoto, Aut '16 14 Tallying versus Multiple Regression

Tallying (Unit Weighting) versus Multiple Regression (MR) Compute cross-validation analyses for Tallying. ○ Fit Tallying Model to 50% of data; predict results for remaining 50% of data. Evaluate the accuracy of prediction. ○ Repeat the analysis for many random 50/50 splits of the data. Compute cross-validation analyses for Multiple Regression. ○ Fit MR Model to 50% of data; predict results for remaining 50% of data. Evaluate the accuracy of prediction. ○ Repeat the analysis for many random 50/50 splits of the data. Which model, Tallying or Multiple Regression produces higher accuracy in prediction? Psych 466, Miyamoto, Aut '16 15 Results of Comparison of Tallying Versus Multiple Regression

Tallying versus Multiple Regression (MR): Which Method has Higher Predictive Accuracy? Czerlinski, Gigerenzer & Goldstein (1999): Averaged over 20 studies, Tallying has (slightly) higher predictive accuracy than MR. Why does this happen? MR overfits the model when the sample size is small. Tallying is robust (gives stable estimates in random data). MR is less robust (its estimates are unstable when sample size is small). Psych 466, Miyamoto, Aut '16 16 GB, Figure 1 (Take-the-Best line has been omitted)

When does Tallying Outperform Multiple Regression (MR)? Tallying outperforms MR when the following criteria are met: 1.Degree of linear predictability is small (R 2 < 0.50; |  | < 0.71). 2.Sample size was less than 10  number of cues; 3.Cues are correlated with each other. Psych 466, Miyamoto, Aut '16 17 Continue This Slide With Additional Comments

When does Tallying Outperform Multiple Regression (MR)? Tallying outperforms MR when the following criteria are met: 1.Degree of linear predictability is small (R 2 < 0.50; |  | < 0.71). 2.Sample size was less than 10  number of cues; 3.Cues are correlated with each other. Conditions 1, 2 and 3 are all associated with increased variability of regression coefficients. Main points: Everyday experience has many correlated cues but not a lot of data. Tallying should outperform MR in everyday experience. Less is more! Psych 466, Miyamoto, Aut '16 18 Transition to Take-the-Best Strategy

Next: "Take the Best" versus Multiple Regression Explain cue validity (key idea in the Take-the-Best strategy) Explain the Take-the-Best strategy Compare Take-the-Best to multiple regression Psych 466,, Miyamoto, Aut '16 19 Explain Cue Validity

Definition of Cue Validity Prediction task: Predicting which of two options has a higher value on a criterion C. Cue validity of feature F = Probability that Option #1 has higher value on C than does Option #2 given that Option #1 has feature F and Option #2 does not. Cue validity of F = P( C Opt.1 > C Opt.2 | Opt #1 has F and Opt #2 does not) Psych 466, Miyamoto, Aut '16 20 Explain: F 1 has greater cue validity than F 2

F 1 has greater cue validity than F 2 : What Does This Mean? Suppose that 1.Option #1 and Option #2 are two options (objects, actions); 2.F 1 and F 2 are two features of these options; 3.C Opt.1 and C Opt.2 are the consequences of choosing Option #1 or Option #2. Then F 1 has greater cue validity than F 2 if P( C Opt.1 > C Opt.2 | Opt #1 has F 1 and Opt #2 does not) > P( C Opt.1 > C Opt.2 | Opt #1 has F 2 and Opt #2 does not) Intuitively, F 1 has greater cue validity than F 2 if knowing that the options differ on F 1 gives you a better chance to guess which option has more of the criterion C than knowing that the options differ on F 2. Psych 466, Miyamoto, Aut '16 21 Describe the Take-the-Best Strategy

"Take the Best" Strategy The Take-the-Best choice strategy has three steps: 1.Examine the cues in decreasing order of their cue validities. 2.Stop as soon as a cue is found on which the options have differing values for the cue. (This is the "first discriminating cue") 3.Take the object that has the higher value on the first discriminating cue. Psych 466, Miyamoto, Aut '16 22 Example of Take-the-Best Strategy

Example of "Take the Best" Strategy Task: Decide whether the Seahawks or Cardinals is more likely to win the Western Division competition. Suppose the cue are: F 1 : Has an excellent defense against the pass; F 2 : Has an excellent defense against the run. F 3 : Has an excellent passing offense; F 4 : Has an excellent running offense Suppose that cue validity of F 1 > cue validity of F 2 cue validity of F 2 > cue validity of F 3 cue validity of F 3 > cue validity of F 4 Psych 466, Miyamoto, Aut '16 23 Continue this Example: Describe the Steps in the Decision

Example of "Take the Best" Strategy (cont.) Take the Best Decision Rule: 1.Choose Team X if Team X has F 1 and Team Y does not. If both teams have F 1 or if neither team has F 1, then proceed to Step Choose Team X if Team X has F 2 and Team Y does not. If both teams have F 2 or if neither team has F 2, then proceed to Step 3. 3.Choose Team X if Team X has F 3 and Team Y does not. If both teams have F 3 or if neither team has F 3, then proceed to Step 2. 4.Choose Team X if Team X has F 4 and Team Y does not. If both teams have F 4 or if neither team has F 4, then guess (make the decision based on a coin flip). Psych 466, Miyamoto, Aut '16 24 Main Feature of "Take the Best": You examine the features in the order from the most to least cue validity. You stop working on the decision as soon as you find a feature that distinguishes between the choices. Results: Take-the-Best versus Tallying versus Multiple Regression

In many cases, Take the Best has greater predictive accuracy than Tallying (Unit Weighting) and MR. Less is more! Take the Best outperforms more complex decision procedures when sample size is small and there are correlations (dependencies) among the cues. Psych 466, Miyamoto, Aut '16 25 Take the Best versus Tallying versus Multiple Regression (MR) Why Do Take-the-Best and Tallying Outperform MR in Prediction

Explaining "Less-Is-More" - Why Is This True (Sometimes)? Decision strategies exhibit a bias/variance dilemma. (a.k.a. bias/variance tradeoff). Psych 466, Miyamoto, Aut '16 26 Graphical Example to Explain the Bias/Variance Dilemma

Example: Bias/Variance Dilemma In Temperature Prediction Larger bias with low variance Better Than Smaller bias with high variance Psych 466, Miyamoto, Aut '16 27 Results for Weather Prediction (GB Figure 3)

GB Figure 3, p. 118 Bias/variance tradeoff for entire year of London temperatures. Psych 466, Miyamoto, Aut '16 28 Claim: Heuristics are Adaptively Superior to More Complex Models

Claim: Heuristics Are Adaptively Superior to Complex Models Humans are better off with biased heuristics that are robust (lower variance) in small samples. Precise normative models are less accurate than heuristic models in small samples. Less-is-more. Reminder: All of the above is only true when the sample size is small relative to the number of predictor variables. ♦ Minimum required sample = 10 x Number of predictor variables (Very approximate formula; it has flaws.) Psych 466, Miyamoto, Aut '16 29 Why Does Heuristics and Biases Research Focus on Decision Errors?

Why Focus on Decision Errors? Make people look stupid. Make ourselves feel smart. Error patterns are clues to cognitive representations and processes. Sometimes errors are important in real life. Psych 466, Miyamoto, Aut '16 30 Amabile (1981), "Brilliant but Cruel." How Should We Test Adaptive Value of Heuristics in the Real World?

Strategy 1: Randomly sample decision. Conceptually and practically, this is hard to do. Strategy 2: Focus on how people make a specific decision. ♦ E.g., Decisions in a hospital emergency room. ♦ E.g., Specific investment decisions. Strategy 3: Study specific decisions in artificial lab settings where variables are easier to control. Psych 466, Miyamoto, Aut '16 31 Final Look at Opposing Camps in JDM Research - END

Heuristics & Biases Research Ecological Rationality Naturalistic Decision Making Ecological Rationality: ♦ Focus on decision environments Naturalistic Decision Making: ♦ Interesting insights, but can they exclude plausible alternative explanations. Heuristics & Biases: ♦ When do heuristics influence important everyday decisions? ♦ What cognitive processes are involved in heuristic reasoning? Memory processes? Attention? Opposing Views in JDM Research Psych 466, Miyamoto, Aut '16 32 END